Structural holes in social networks

Sanjeev Goyal∗ Fernando Vega-Redondo†

This version: December 2004

Abstract

We consider a setting where every pair of players that interact (e.g. exchangegoods or information) create a surplus. An interaction can take place only if theplayers involved have a connection. If the connection is direct the two playerssplit the surplus equally while if it is indirect then intermediate players also geta share of the surplus. Thus individuals form links with others to create surplus,to gain intermediation rents and to circumvent others who are trying to becomeintermediary.Our principal result is that strategic link formation in such a setting leads to thestar network. In a star a single agent acts as an intermediary for all transactionsand there is significant payoff inequality across ex-ante identical players.

∗Department of Economics, University of Essex, sgoyal@essex.ac.uk.†Department of Economics, Universidad de Alicante, University of Essex, and Instituto Valenciano de

Investigaciones Economicas, vega@merlin.fae.ua.es.We have benefitted from comments by Coralio Ballester, Andrea Galeotti, Suresh Mutuswami, and ArnoRiedl, and from presentations at Warwick, Stockholm, Oxford, and Barcelona.

1

1 Introduction

It is now widely agreed that knowledge of the structure of interaction among individuals

is important for a proper understanding of a number of important questions in economics,

such as the spread of new ideas and technologies, the patterns of employment and wage

inequality, competitive strategies in dynamic markets, and career profiles of managers.1

Connections facilitate timely access to important information – on trade opportunities,

job vacancies, project deadlines, and novel ideas for research. In some important instances

– e.g., trade opportunities, the combining of novel research ideas – the payoffs an indi-

vidual entity gets in a network will clearly depend on his relative importance in bridging

gaps in the network between others.2 The potential gains from bridging different parts of

a network were important in the early work of Granovetter (1974) and are central to the

notion of structural holes developed by Burt (1994). In recent years, a number of empir-

ical studies have shown that individuals or organizations who bridge ‘structural holes’ in

networks gain significant payoff advantages.3 For instance, the work on promotions and

performance evaluation argues that the differences in structural location of individuals

– in particular whether they bridge structural holes in the social network – explains a

significant part of the variation in promotion timing of otherwise similar people. Given

these significant payoffs effects, it seems natural that an individual will make invest-

ments in connections so as to become structurally important, while other individuals will

likewise form connections to circumvent such attempts. Are special structural positions

and the corresponding large payoff differences sustainable when individual entities form

connections strategically?

1We mention a small sample of this work. See Bala and Goyal (1998), Coleman, Katz, and Menzel(1966) and Ellison and Fudenberg (1992) for the role of communication networks in technological diffusion;Rees (1966), Granovetter (1974), Montgomery (1992) and Calvo and Jackson (2004) for effects of socialnetworks on employment and wage inequality; Ahuja (2000), Goyal and Moraga (2001) and Gulati, Nohriaand Zaheer (2000) on the role of strategic partnering in generating competitive advantage in markets;Burt (1994) for the effects of organizational networks on individual promotions and salaries, and Rauch(2001) on the role of social and economic networks in internation trade.

2Burt (2004) explores the influence of individual position in social networks in shaping the generationof creative ideas.

3See Burt (1994) and Mehra, Kilduff and Brass (2003) for influence of structural positions on promo-tions and performance evaluation, Podolny and Baron (1997) for work on network positions and mobility,and Ahuja (2000) for the influence of a firm’s position in inter-organizational networks on its innovative-ness and overall performance.

2

We develop a simple model of network formation to address this question. We consider

a setting where interaction between every pair of individuals generates a surplus. If

two individuals are directly linked then they split this surplus equally, while if they are

indirectly connected – there are other players in the ‘path’ between them – then the

division of surplus depends on the competition between these intermediaries. In this

setting, there are three types of incentives for individuals to form links with others. The

first incentive is the desire to create surpluses: individuals would like to join the network

so as to create exchange possibilities which in turn create surpluses. The second incentive

is related to the rewards from intermediation: players would like to place themselves

between others in order to extract rents from intermediation. The third incentive arises

out of the desire to avoid sharing surpluses with intermediaries; in other words, individuals

will try to circumvent intermediate players to retain more of the surplus for themselves.

Our principal finding is that strategic link formation leads to the star network. This is

a network in which one player (referred to as the central player) forms links with all the

other players (who are referred to as the peripheral players) and there are no other links

in the network. There are two aspects of the result that we would like to stress: one, the

star is the unique non-empty network that arises in equilibrium and two, the star entails

significant payoff inequality among ex-ante identical players. Thus an extreme version

of strategic positioning – with the central player earning significantly larger payoffs – is

the only possible outcome in a setting with ex-ante identical individuals. We now briefly

sketch the main arguments underlying this result.

The first observation is that, due to the role played by access benefits – the surplus gen-

erated from direct and indirect interaction– an equilibrium network must be either empty

or connected. Thus, among non-empty networks, we only need to check which connected

networks are robust with respect to individual incentives. The second observation is that

if the network is minimally connected (i.e. has no cycles) then long paths cannot be

stable. This is because players located at the ‘end’ of the network benefit from connect-

ing to a “central player” in order to save on intermediation costs (cutting path lengths)

and, on the other hand, a central player is also ready to incur the cost of an additional

link because this enhances her intermediation payoffs. Thus there is a natural tendency

for minimally connected networks to become agglomerated around central players. But

what about networks with cycles? In principle, a cycle containing all players appears to

be very attractive. Every player is in a symmetric position and each of them can access

3

every other. Such a cycle network, however, is vulnerable to the incentives of individuals

towards becoming uniquely intermediate and reaping the entailed intermediation benefits.

To see this, consider two players that are far apart in the cycle and establish a direct link.

By simultaneously breaking one link each, they can produce a “line” and become central

in it. In a line, they must pay intermediation costs to a number of others, but their promi-

nent centrality more than offsets these costs through even larger intermediation benefits.

A similar argument can be used to break any network with cycles, thus leaving the star

network as the only (non-empty) equilibrium network.

We examine the robustness of these arguments in a number of directions. In the basic

model intermediation rents arise only if a player is essential. We consider an alternative

formulation where intermediation rents increase smoothly in the extent of the criticality

of a player and find that stars arise in that setting as well. In the basic model stars

arise in societies with a large number of players. This leads us to examine the nature

of networks when individuals are constrained in the number of links they can form. We

find that these constraints give rise to networks with a group of inter-connected central

players who are local stars each with their own distinct group of peripheral players.

We now clarify the contribution of our paper by discussing its relationship with research

in economics, sociology and the literature on complex systems. In recent years, much

attention has been devoted by economists and game theorists to the theory of network

formation.4 In existing work it is assumed that the goods transacted via the links are

non-rival — see e.g. the connections model studied by Jackson and Wolinsky (1996) and

the version with one-sided link formation proposed by Bala and Goyal (2000). Here we

consider a setting with rival goods and explicitly incorporate the idea of intermediation

rents and payments. The allocation of intermediation rents in turn requires a model of

competition between intermediaries and a related contribution of our paper is a method

to assign these benefits as a function of the network structure. We introduce the idea of

essential players – players without whom an interaction cannot take place – to determine

which intermediaries will earn rents. The incorporation of intermediary rents and pay-

ments has powerful implications for equilibrium networks. For example, in earlier work

on the two-sided connections model, stars can be sustained only over a small range of pa-

rameter values, and the central player who is bridging the structural holes actually earns

4See Aumann and Myerson, (1989), Boorman (1975), Bala and Goyal (2000), Dutta, Tijs and van denNouweland (1998), Jackson and Wolinsky (1996) and Kranton and Minehart (2001).

4

a lower payoff than the peripheral players (see Proposition 2 in Jackson and Wolinsky,

1996). By contrast, in the present model, the star is sustainable for practically the entire

range of parameters and the central player earns a much higher payoff as compared to

the peripheral players.5

The ideas of access advantages and strategic positioning are important building blocks for

the notion of structural holes, introduced by Burt (1994), and have been an important

part of the tradition in sociology since the work of Granovetter (1974). As we mentioned

earlier, empirical research by sociologists shows that differences in structural location

of otherwise similar individuals – in particular whether they bridge structural holes in

the social network – explains a significant part of this variance. Our paper shows that

the presence of structural holes and corresponding payoffs differences is consistent with

a model of rational players who strategically seek to create positional advantages for

themselves and also have an incentive in preventing others from becoming central. To

the best of our knowledge this is the first formal model to develop an explanation for

structural holes based on intermediation rents and the large inequality in payoffs that go

with them.

In recent years statistical physicists studying complex networks have documented a num-

ber of empirical properties – small average distances, high inequality in number of links

across nodes of the network, among others – of large social, biological and technological

networks.6 One of the leading models in this literature is that proposed by Barabasi and

Albert (1999). They formulate a simple dynamic model of network formation in which,

at every point in time, a new node arrives and forms new links with existing nodes. The

probability that these links connect to any given existing node is taken to be increasing

in –typically, proportional to – the number of links the node has; this is referred to as

preferential attachment. The property of preferential attachment is critical for the deriva-

tion of unequal link distributions in the long run – in particular, the so/called scale-free

5Recent papers by Feri (2004) and Hojman and Szeidl (2004) also explore the emergence of starnetworks. Their approach, however, is different from the one pursued here in two important respects.First, as in the connections model, the transacted goods are non-rival; second, links are one-sided (i.e.they are fully payed by the player who initiates them). It is precisely this second feature that allows forthe establishment of periphery-sponsored stars in large populations: all links are initiated by peripheralplayers and thus the center incurs no cost at all for any of her links. Another model where stars mayarise at equilibrium is studied by Galeotti and Melendez (2004). Their approach, however, has agentsinvolved in an infinitely repeated Prisoner’s Dilemma that determines how linking costs are shared.

6For a survey of this work see Albert and Barabasi (2002), Newman (2001), and Vega-Redondo (2005).

5

(or power-law) degree distributions. This literature, however, does not provide us with

any reasons for why a highly linked older player and a newly arriving player should want

to connect with each other. We show that large intermediation rents for the older player

and the desire of the new player to minimize the number of intermediaries (and thereby

reduce payments) jointly provide a simple explanation for preferential attachment.

The paper is organized in five sections. Section 2 lays out the basic model. Section 3

presents the main results on equilibrium and efficient networks. Section 4 discusses the

role of different assumptions in our analysis, while section 5 concludes. All the proofs are

given in an Appendix at the end of the paper.

2 The Model

We consider a population composed of finite set of ex-ante identical agents, N = {1, 2, ..., n}where n ≥ 3. These agents play a network-formation game where every one of them

makes a simultaneous announcement of intended links. An intended link si,j ∈ {0, 1},where sij = 1 means that player i intends to form a link with player j, while sij = 0

means that player i does not intend to form such a link. Thus a strategy of player i is

given by si = [sij]j∈N\{i}, with Si denoting the strategy set of player i.

A link between two players i and j is formed if and only if sij = sji = 1. We denote the

formed (undirected) link by gi,j ≡ gj,i = 1 and the absence of a link by gi,j ≡ gj,i = 0. Any

given strategy profile s = (s1, s2, ..., sn) therefore induces a network g(s). The network

g(s) = {(gi,j)}i,j∈N is a formal description of the pair-wise links that exist between the

players. There exists a path between i and j in a network g if either gi,j = 1 or if there

is a distinct set of players {i1, . . . , in} such that gi,i1 = gi1,i2 = gi2,i3 = · · · = gin,j = 1.

All players with whom i has a path defines the component of i in g, which is denoted by

Ci(g).

We shall suppose that any pair of players j and k who are connected by a path generate a

unit of surplus. The allocation of this surplus depends on whether there are intermediaries

between j and k and on the competition between the intermediaries. In the basic model we

will take the view that any two paths between any two players j and k fully compete away

all the surplus. This is in the spirit of Bertrand competition between paths. In general,

the unit of surplus may be divided among all agents in N according to some imputation

zjk = (zjki )i∈N of non-negative shares. In the Appendix we show that the above payoff

6

division is the unique allocation in the kernel of the corresponding cooperative game (see,

Davis and Maschler (1965)).

We are then led to the notion of essential players: a player i is said to be essential for j

and k if i lies on every path that joins j and k in the network. For example, the number

of essential players between j and k is zero if the players have a direct link, or if players

are located around a circle – in the latter case, for every pair of players j and k and every

other player i, there is always a path joining j and k that does not include i. On the

other hand, note that in a star every pair of peripheral players has a single and common

essential player, namely the center of the star.

We suppose that agents have to pay a fixed (marginal) cost c for each the links they

establish. Denote by E(j, k; g) the set of players who are essential to connect j and

k in network g and let e(j, k; g) = |E(j, k; g)|. Then, for every strategy profile s =

(s1, s2, ..., sn), the (net) payoffs to player i are given by:

Πi(si, s−i) =∑

j∈Ci(g)

1

e(i, j; g) + 2+

∑

j,k∈N

I{i∈E(j,k)}e(j, k; g) + 2

− ηi(g)c, (1)

where I{i∈E(j,k)} ∈ {0, 1} stands for the indicator function specifying whether i is essential

for j and k, and ηi(k, g) ≡ |{j ∈ N : j 6= i, gij = 1}| denotes the number of players with

whom player i has a link.

We note that even if a player lies on several paths between two players she may still get

no intermediation payoffs if she is not essential. In Section 4 we examine in detail the

case where a player i’s intermediation rents from connecting j and k are an increasing

function of the number of shortest paths (between j and k that) she lies on. We also note

that the payoff function above assumes that the costs of link formation are constant. In

section 4 we study the implications of increasing costs by considering the case of capacity

constraints on the number of links an individual can form.

The main objective of the paper is studying the architecture of networks that are strate-

gically stable and assess their efficiency. Our notion of strategic stability is a refinement

of Nash equilibrium that allows for coordinated two-person deviations.

Definition 1 A strategy profile s∗ is a Bilateral Equilibrium (BE) if the following con-

ditions hold:

• for any i ∈ N, and every si ∈ Si, Πi(s∗) ≥ Πi(si, s

∗−i);

7

• for any pair of players i, j ∈ N , and every strategy pair (si, sj),

Πi(si, sj, s∗−i−j) > Πi(s

∗i , s

∗j , s

∗−i−j) ⇒ Πj(si, sj, s

∗−i−j) < Πj(s

∗j , s

∗j , s

∗−i−j).

We apply the term ‘bilateral equilibrium’ to a strategy profile where no player or pair of

players can deviate (unilaterally or bilaterally, respectively) and benefit from the deviation

(at least one of them strictly, for bilateral deviations). This notion refines the original

formulation of pair-wise stability due to Jackson and Wolinsky (1996) by allowing pairs

of players to form and delete links simultaneously.

In fact, our analysis will focus on a refinement of BE that we call strict. It rules out the

existence of deviations (unilateral or bilateral) that have some consequence (i.e. affect

the network) but nevertheless are payoff indifferent for the agents involved. In part,

the motivation of this equilibrium concept is dynamic: a gradual adjustment process

that allows pairs of players to revise the network in sequence will (only) be absorbed by

equilibria of this sort – see below for an elaboration.

Definition 2 A strategy profile s∗ is a Strict Bilateral Equilibrium (SBE) if the following

conditions hold:

• for any i ∈ N , and every si ∈ Si such that g(si, s∗−i) 6= g(s∗), Πi(s

∗) > Πi(si, s∗−i);

• for any pair of players, i, j ∈ N and every strategy pair (si, sj) with g(si, sj, s∗−i−j) 6=

g(s∗),

Πi(si, sj, s∗−i−j) ≥ Πi(s

∗i , s

∗j , s

∗−i−j) ⇒ Πj(si, sj, s

∗−i−j) < Πj(s

∗j , s

∗j , s

∗−i−j).

We note that a strict bilateral equilibrium is a bilateral equilibrium.

Finally, we introduce the notion of efficiency. In line with the assumption that the bar-

gaining setup involves transferable utility, different networks g are assessed in terms of the

total surplus generated, W (g) ≡ ∑i∈N Πi(g). Let G denote the set of all possible networks

(i.e. all undirected graphs with n vertices).

Definition 3 A network g is efficient if W (g) ≥ W (g) for all g ∈ G.

8

Before undertaking the analysis of the model, it might useful to review some standard

graph-theoretic notions that will be used repeatedly. A network is said to be connected

if there exists a path between any pair i, j ∈ N . Given any g′ ⊂ g, let N(g′) ≡ {i ∈ N :

g′ij = g′ji = 1 for some j} be the subset of nodes which display some link in g′. Then, a

network, g′ ⊂ g, is a component of g if for all i, j ∈ N(g′), i 6= j, there exists a path in

g′ connecting i and j , and for all i ∈ N(g′) and k ∈ N , gi,k = 1 implies k ∈ N(g′). A

component g′ ⊂ g is complete if gi,j = 1 for all i, j ∈ N(g′).

Two networks g and g′ are said to have the same architecture if one network can be

obtained from the other by a permutation of the players’ labels. A network is said to be

symmetric if all players have the same number of links, say η. The complete network, gc,

is a symmetric network in which η = n − 1, ∀i ∈ N , while the empty network, ge, is a

symmetric network in which η = 0, ∀i ∈ N . Another example of a symmetric network is

a cycle where η = 2 and the whole set of nodes can be ordered in a list i1, i2, ...in with

gi1,i2 = gi2,i3 = .... = gin,i1 = 1.

Finally, we shall say that a network is asymmetric if there is at least one pair of players

who have a different number of links. One important example is the star where there is

a single node, ic, with ηic = n− 1 while ηi = 1 for all other i 6= ic. An interesting mixture

is what we shall call the hybrid cycle-star network where there is a subset of k nodes

C = {i1, i2, ...ik} arranged in a cycle and some particular player ix ∈ C such that gj,ix = 1

for all j ∈ N\C. Figure 1 illustrates these different types of networks.

3 Analysis

Our main result is that equilibrium networks must be stars. This result shows that

structural holes arise endogenously in the starkest possible manner – a single individual

(the center of the star) is essential to the value generated in the whole network. This in

turn implies that the center also appropriates a correspondingly large part of the total

surplus generated by the network.

We start by noting a property of equilibrium networks.

Proposition 1 A bilateral equilibrium network is either empty or connected.

9

Proof: See Appendix.

Suppose that, contrary to what is asserted, an equilibrium network is split into two or

more components. Consider two individuals, i and j, in different components. First, we

observe that their marginal payoffs of establishing a link between them (thereby merging

the two components) are exactly the same for both players. This happens because the

additional “access payoffs” of i from connecting to the individuals in the other component

are identical to the corresponding intermediation payoffs earned by j. Therefore, since

gross payoffs are the sum of access and intermediation payoffs, both players enjoy the

same marginal gross benefit from linking to each other. Now, we argue that it cannot

be optimal for these players to remain in separate components. The reason is as follows.

Suppose that i’s component contains more than one player. We show that the component

must have some agent (let us suppose that it is i herself) that enjoys no intermediation

payoffs (either because she is extremal or an “inessential” part of a cycle). Then, the

payoffs of j from linking with i are, in addition to the payoff from the direct link, a

“scaled down replica” of the access payoffs of i. It then follows that the gross payoffs to

j must exceed the linking cost if, as assumed, it is optimal for player i to incur the cost

of two links to have her (pure access) payoffs.

What type of equilibrium networks are then possible? We start by noting that two types

of architectures are always possible, unless the cost is very low or the population is quite

small. On the one hand, the empty network is SBE if c > 1/2, even if the population is

arbitrarily large. This reflects a simple instance of “coordination failure”: if no links are

formed, then the creation of any link has to be judged on a stand-alone basis, which is

unprofitable if the linking cost exceeds half of the unit surplus earned by an isolated pair

of players.

On the other hand, it is also clear that the star is a SBE if the population is large enough

and the linking cost is no so low as to justify a direct connection with every other player.

Specifically, suppose that 1/6 < c < 1/2 + (n − 2)/6. Then, in a star, the payoffs to the

center are positive and equal to

n− 1

2+

1

3

(n− 1)(n− 2)

2− (n− 1)c

whereas if she were to keep only k of the n− 1 links she would earn the lower payoffs:

k

2+

1

3

k(k − 1)

2− kc.

10

As for the incentives of the peripheral players, in the star they earn 1/2+(n−2)/3−c > 0

while if they created an additional link the entailed payoffs would be 1/2 + 1/2 + (n −3)/3− 2c. The latter is smaller than the former if c > 1/6.

The star is a (strict) bilateral equilibrium network (SBE) for a wide range of parameters

and this is due to the fact that centrality generates large payoffs from essentialness.

However, a star also exhibits an extreme form of such essentialness, with only one player

being essential for all pairs of players. This raises the question: are there other – perhaps

more egalitarian – network architectures that can arise in equilibrium? The following

Theorem, the main result of the paper, provides a complete (negative) answer to this

question for large societies.

Theorem 1 Suppose that n is large. If 1/6 < c < 1/2 then the unique SBE network is

the star, while if 1/2 < c < 1/2 + (n− 2)/6 then the star and the empty network are the

only SBE networks. If c > 1/2 + (n− 2)/6 then the empty network is the unique bilateral

equilibrium network.

Proof: See Appendix.

The proof presented in the Appendix proceeds by showing that all networks other than

the empty network and the star are not sustainable in a strict equilibrium. The arguments

rely on the three kinds of incentives that arise in our model: accessing others, gaining

intermediation rents, and avoiding intermediation payments. The intuition underlying

the main steps can be outlined as follows.

1. If a SBE network is minimally connected (i.e. a tree), it must be a star

This follows from two observations. First, the number of essential players between

any two players is bounded above, independently of the population size (if it were

too large, the incentives of these players to establish a link and thus gain shorter

access to each of them as well as to many others would certainly exceed the cost

of the link). Secondly, “extremal” and “central” players always have much to gain

by connecting directly since by so doing the gains are proportional to population

size but (opportunity) costs are bounded. A simple illustration of the latter point

is depicted in Figure 2.

11

2. There can be at most one cycle in a SBE network

Here, the main point is that, if several cycles exist in a network, it is always possible

for two players to establish a new link and at least remain equally well off. Two

possible cases need to be distinguished in this respect, illustrated in turn by Figures

3A and 3B. In the first case, the two cycles have no players in common and players

such as 3 and 4 in Figure 3A.1 have a strict incentive to connect and destroy their

links to essential players 1 and 2. This change leads to the network in Figure 3A.2,

where players 3 and 4 access the same number of other players, avoid the need of

sharing some payoffs with players 1 and 2, and incur the same linking costs. In the

second case, there are common players in both cycles (i.e. player 1 in Figure 3.B.1),

so that a deviation by two players (1 and 2) can transform those two cycles into just

one (Figure 3.B.2) where the payoffs to both of them are exactly the same as before.7

This means that the original network fails to be a strict bilateral equilibrium.

3. A single cycle cannot be a SBE network

A cycle provides incentives for players lying on opposite sides of it to connect and,

by also deleting two of their links, become markedly central and thus earn large

intermediation rents (cf. Figure 4). It is true that by so doing the two players

in question have to make intermediation payments to others where none of these

existed under the cycle. We find however that the intermediation rents dominate

the intermediation payments and as a result the cycle is not sustainable.

4. If a SBE network includes a cycle, all other players not belonging to it are connected

to the same player in the cycle, i.e. it is a hybrid cycle-star network

The starting point of the argument is the observation that, given the linking cost c,

the cycle cannot include more than a certain number of players – otherwise, creating

a link that bypasses the “gate player” that is essential to accessing the cycle would

become profitable. Then, since almost all players (if the population is large) must

lie outside the cycle, we can rely on an adaptation of the ideas used in Step 1 above

to argue that, in the minimally connected part of the network that lies outside the

cycle, players must be arranged in a star configuration.

7This change in link pattern, however, strictly increases the payoffs of player 3 since player 3 has thesame access benefits, no essential players, but forms one link less in the new network.

12

5. The only hybrid cycle-star network that defines a SBE is the “degenerate” star net-

work

As explained above, in a SBE network that involves a unique cycle most of the

players in the (large) population must lie outside the cycle and connect to it through

a single player. Therefore, much as it was argued for Step 3, that configuration opens

the possibility that the cycle be broken by the concerted deviation of two players in

it, who can then obtain high intermediation payoffs. The nature of such a deviation

is illustrated in Figure 5. It rules out that a hybrid cycle-star SBE network may

include any genuine cycle, thus leaving us with the strict star network as the only

possibility for a non-empty SBE. Since it is clear that a star network is a SBE (see

the discussion preceding the statement of the theorem), the conclusion follows

The above result provides us with a complete characterization of equilibrium networks

for large societies. What can we say about equilibrium networks in small societies? We

first note that a star is an equilibrium for small societies so long as 1/6 < c < 1/2 + (n−2)/6, and that any minimal network contains structural holes and corresponding payoff

inequality. Second, we note that steps 2 and 3 do not use the size of society, and so a

single cycle or a network with several cycles cannot arise even in small societies. Finally,

step 4 shows that, in a network with one cycle and some players outside it, there is an

upper bound to the number of players in the cycle. In other words, most of the players

are peripheral to a ‘local’ star. These considerations lead us to conclude that equilibrium

networks in small societies also have a tendency to generate structural holes and there

will be players who will earn large intermediation payoffs by spanning them.

We end this section by discussing briefly the architecture of socially optimal networks.

First, we observe that from a social point of view there is no gain in adding links in a

connected network, while there is a cost to adding links since c > 0. So efficient networks

must be minimal or empty. On the other hand, in our setting where positive surpluses

are earned from every direct and indirect connection, it is easy to see that networks with

distinct (non-trivial) components cannot be efficient. Thus, an efficient network must

either be empty or minimally connected. Finally, we note that the aggregate net payoff

in the latter case is 12n(n− 1)− 2c(n− 1), which is positive if, and only if, c < n/4. Based

on these observations, the following result readily follows.

13

Proposition 2 If c < n/4 then an efficient network is minimally connected, while if

c > n/4 then an efficient network is empty.

Clearly, the star is a minimally connected network and is therefore efficient for large

n. Therefore, combining Proposition 2 and Theorem 1, we conclude that efficiency is

guaranteed at a nonempty SBE network, provided the population is large.

4 Discussion

We have shown that in a setting with access benefits and intermediation rents strategic

network formation generates sharp predictions: for large societies, all non-empty equilib-

rium networks are stars. In this section we will examine how sensitive this result is to

some of the specific assumptions we made in the analysis. First, In Subsection 4.1, we

explore a variant of the model where payoffs depend on the length of connecting paths

rather than on the existence of competing alternatives. Second, in Subsection 4.2, we

discuss the implications of capacity constraints in the number of links that each player

can individually support. Finally, in Subsection 4.3, we check whether our analysis is

robust to the requirement of credibility “coalition-proofness” in bilateral deviations.

4.1 Essentiality and intermediation

In the model we have assumed that a player gets intermediation payoffs if and only if she

is essential. This formulation has intuitive appeal in terms of a form of “Bertrand-like”

competition between paths connecting individuals. In fact, it is such form of competition

that underlies the intuition that any allocation in the Core gives zero payoffs to non-

essential players, and it is also partly reflected in the selection of the Core afforded by

the Kernel. In some settings, however, it may be argued that a player’s intermediation

payoffs should depend on the nature of the connecting path on which she lies – i.e. not

only on whether there are some alternatives. For example, if we abstract from inter-path

competition, a natural assignment rule would be one prescribing that all players in the

path actually used to connect any two players are to receive an equal share of the surplus.

Indeed, this is the outcome to be expected if the decision of what path to use must be

determined ex ante in an irreversible manner. In that case, once such a decision has been

taken, the existence of alternative paths should play no role in the assignment of the

surplus and all players involved should receive an equal share.

14

Here, we discuss briefly the implications of this alternative approach. Note that an im-

mediate implication of it is that only shortest paths will be used. That is, any player i

aiming to connect to j will always choose to start the endeavor by contacting a direct

neighbor that lies on one of the shortest path to j. Then, subsequently, since all other

players in that path share the same objective, a shortest path will indeed be used. This

induces payoff functions Πi that, for any strategy profile s = (s1, s2, ..., sn), are defined as

follows:

Πi(s1, s2, ..., sn) =∑

j∈Ci(g)

1

d(i, j; g) + 1+

∑

j,k∈Ci(g)

bi(j, k; g)

d(j, k; g) + 1, (2)

where d(i, j; g) denotes the (geodesic) distance between i and j in network g, whereas

bi(j, k; g) stands for the fraction of shortest paths in g joining j and k that pass through

i.

We now argue that many of the insights obtained in the original model carry over to

this setting. For simplicity, let us restrict attention to the collection of networks where

the tension between symmetry and polarization arises in a simple but stark manner:

the hybrid cycle-star networks. These networks include the pure cycle – reflecting no

intermediation rents – on one end, and pure stars – reflecting extreme intermediation

power – on the other end. Within the original model, the tension was resolved in favor of

extreme polarization (i.e. stars). We claim that the same occurs as well for the present

version where payoffs are given by (2).

To see this, consider any hybrid cycle-star network where there are m peripheral players

connected to a single player i in a cycle, which consists of the remaining n −m players.

We now show that this network cannot be sustained at a SBE if m < n, i.e. whenever

the network “falls short” of being a pure star. (Note that this statement covers the cycle

as well, which arises when m = 0.) Let j be the player opposite to i in the cycle. Suppose

that i and j are given the opportunity of forming a link. We claim that they will always

form it, since they can profit from so doing if they also delete one link each and transform

the network into a hybrid line-star network with i and j central. (Here, the deviation is

akin to that illustrated in Figure 4 concerning the instability of the cycle in the original

model.) This follows from the following observations:

1. Player i has the access benefits unchanged, i.e. it accesses the same number of

players with correspondingly the same number of intermediate players as before.

15

2. Player j has weakly more access benefits than before: she accesses the same number

of players as before and (if m > 0) the peripheral ones with fewer intermediaries

(now, only player i).

3. Player i has more intermediation payoffs than before: she keeps matters unchanged

concerning the connections between agents in the line and the peripheral players,

but she earns more intermediation payoffs than before by connecting agents in the

line (she is now relied upon for more bilateral connections).

4. Player j has the same intermediation payoffs as before concerning connections be-

tween individuals in the line but has additional such payoffs (whereas before she had

none) concerning connection of some agents in the line and the peripheral players.

Since the number of links by players i and j do not change by the deviation, 1-4 implies

that the bilateral deviation considered is jointly profitable. This shows that none of

the networks considered other than the pure star is sustainable at a SBE. This suggests

that in the alternative model proposed the same tendency towards polarization arises,

and again materializes in an extreme fashion, within a parametrized family of networks

i.e. the collection of hybrid cycle-star networks. Finally, we note that an analogous

insight is gained if we focus our attention on minimally connected networks (i.e. trees).

Circumscribed to these networks, the two payoff functions (Πi and Πi, as given respectively

by (1) and (2)) coincide, which implies that Point 1 in the discussion following Theorem

1 (i.e. Lemma 3) applies unchanged, i.e. among the minimally connected networks, all

SBE are stars.

4.2 Capacity constraints

In the basic model we assumed that the marginal costs of linking between players are

constant and in particular do not depend on the number of links a player forms. In some

settings it seems more natural to suppose that the costs per link are increasing in the

number of links. Or, analogously, we should expect that an individual player will be

subject to some capacity constraints. In this section we briefly discuss the implications

of such constraints on equilibrium networks.

Let us suppose that any individual can form a maximum of K << n − 1 links. This

capacity constraint means that the agglomeration forces at work in Theorem 1 are now

constrained. How does this affect the arguments underlying this result? Reconsidering for

16

this case the logic organized in Points 1-5 above (cf. the discussion following the statement

of Theorem 1, as well as the lemmas in the Appendix), we arrive at the following set

of observations. First, note that the forces towards agglomeration spelled out in Point

1 (Lemma 3) still operate unchanged up to the point where capacity constraints are

reached. Next, we observe that the arguments used in Points 2, 3 and 5 (lemmas 4, 5,

and 7) do not involve players forming any additional links and therefore carry over to

a setting with capacity constraints. This means that networks with two or more cycles,

networks with a single cycle, or a hybrid cycle-star network cannot arise in equilibrium.

Therefore, the only possibility that remains is a network with a cycle and some players

outside this cycle. Of course, many different network architectures are still consistent

with this description. But all of them must display significant agglomeration polarized at

some players, as allowed by their capacity constraints. More specifically, one expects that

equilibrium networks include a cycle composed of highly connected capacity-constrained

individuals, who act as local hubs and are also the gates through which peripheral players

(outside the cycle) access most of the population. In the end, this suggests that, even

when players are capacity constrained, high centrality and significant payoff inequality

still arise as important features of equilibrium networks.

4.3 Two-person coalition proof networks

In our analysis so far we have assumed that a deviation by two players is credible so long as

it yields higher payoffs to both players. We have not looked at profitable deviations from

the agreed upon deviation, thus ignoring considerations that are in the spirit of coalition

proofness. In this section we will examine the implications of this further requirement,

which of course can only enlarge the set of equilibria – in general, only a subset of bilateral

deviations qualify as valid. We start with a definition of bilateral proofness in our context,

focusing directly on the “strict” version of this concept.

Definition 4 A strategy profile s∗ is a Strict Bilateral-Proof Equilibrium (SBPE) if the

following conditions hold:

1. for any i ∈ N , and every si ∈ Si such that g(si, s∗−i) 6= g(s∗), Πi(s

∗) > Πi(si, s∗−i);

2. For any pair of players i, j ∈ N, and every strategy pair (si, sj) with g(si, sj, s∗−i−j) 6=

g(s∗) and Πi(si, sj, s∗−i−j) ≥ Πi(s

∗i , s

∗j , s

∗−i−j), one of the following two conditions

hold:

17

(a) Πj(si, sj, s∗−i−j) < Πj(s

∗j , s

∗j , s

∗−i−j)

(b) ∃k, ` ∈ {i, j}, k 6= `, and some sk ∈ Sk, such that Πk(sk, s`, s∗−k−`) > Πk(sk, s`, s

∗−k−`).

Most of the insights obtained in Section 3 through the SBE concept are maintained if

we consider the less demanding SBPE concept. Indeed, reviewing the Propositions and

Lemmata that underlie the proof of Theorem 1, it is straightforward to check that all of

them continue to apply for the SBPE concept, except for Lemmas 5 and 7 (corresponding

to Points 3 and 5 in the discussion following Theorem 1). In particular, it is possible to

show that a cycle and hybrid cycle-star networks can be sustained in strict bilateral proof

equilibrium. The following result shows it for the cycle.

Proposition 3 Given any c > 0, there exists an n(c) such that for all n ≥ n(c), a cycle

containing all players is a strict bilateral proof equilibrium.8

The proof of this result is presented in the Appendix. It shows that the deviation by two

distant players in a cycle which takes the cycle to a line is not credible since each of the

players has an incentive to renege on the deviation and retain both their erstwhile links,

under the assumption that the other player will delete one her links. This incentive is

clarified in Figure 6. Recall that, in Theorem 1 (cf. Point 4 following its statement), a

cycle was broken by a deviation in which, say, players 1 and 2 in Figures 6A and 6B move

and create a line. The proof of Proposition 3 shows that this deviation is vulnerable to

a further deviation in which player 1 retains both the links that she had. This deviation

yields network 6C. By moving to this network, player 1 is able to circumvent a large

number of essential players in the line. This deviation is worthwhile if there are enough

players on the line going from 1 to 3. An analogous reasoning can be used to discard

bilateral deviations from a hybrid cycle-star network: a joint deviation by two players

in the cycle is not in the interest of any one of them, under the assumption that the

other abides by it. In fact, it is not difficult to construct examples (see Remark 1 in the

Appendix) where a hybrid cycle-star network is SBPE for a large enough population.

The above discussion indicates that, under the requirement of bilateral proofness, a richer

set of network architectures can be supported at equilibrium – specifically, the cycle and

a hybrid cycle-star network are also possible, in addition to full stars. Thus, in this sense,

the force towards polarization that proved so acute under unqualified bilateral stability

8The stability of the cycle (which is not minimally connected) is an instance of the inefficiency discussedin Jackson (2003) where players over-connect to avoid positional disadvantages.

18

becomes mitigated when we insist that bilateral deviations be internally consistent. We

now argue, however, that such alternative architectures (a cycle and a hybrid cycle-star)

are, in effect, rather fragile. Specifically, they should not be expected to last under the

pressure of occasional perturbations that affect some of the links. A natural way of

formulating precisely this idea is to check the performance of some suitably modelled

dynamics of adjustment under the pressure of occasional random decay of links. In what

follows, we outline such a dynamic model and explain the gist of the argument. While we

dispense with the formal details, they are available from the authors upon request.

Consider a dynamic process of bilateral adjustment in which, at every point in (discrete)

time t = 1, 2, ..., a pair of players is selected at random and are given the option to create

a link between them (if this link is not already in place) and, simultaneously, destroy any

subset of the existing links whose maintenance they control. In line with the considerations

that underlie the SBPE concept, let us suppose that, in evaluating any joint move by

the two players involved in the adjustment, only those that are internally consistent are

admitted as valid. That is, joint moves are only judged valid if they are immune to a

unilateral deviation (that would always involve keeping a link that should otherwise be

maintained under the contemplated move). In addition to this adjustment dynamics,

suppose that there is also a “slow” dynamics of link decay. Specifically, postulate that, at

(the end of) every t, each of the existing links vanishes with some probability ε, conceived

as small.

The dynamics just described may be parameterized by ε. If ε = 0, we simply have the

pure adjustment dynamics, whose stationary points are SBPE networks – they include

cycle, hybrid cycle-star, or star networks, together with the empty network if and only

if c > 1/2. To check the robustness of these different architectures, suppose ε > 0

and consider what would happen if one of the existing links, randomly selected, simply

vanishes. The implicit assumption here is that such a perturbation is very infrequent so

that no more than one instance of it may occur with significant probability before the

adjustment dynamics restores the stationarity of the situation. For concreteness, we focus

our discussion on the cycle, since the arguments pertaining to a hybrid cycle-star network

are very similar.

After a link in the cycle has vanished, a line network obtains. We argue that, thereafter,

there is positive probability that the adjustment dynamics leads to a star, which is not only

19

absorbing for this dynamics but robustly so. To show positive probability of convergence,

it is enough to construct a finite admissible path of adjustment that leads to a star. The

details are somewhat involved, and spelled in some detail in the Appendix (cf. Remark

2). The main idea, however, is simple, being again a reflection of the forces towards

agglomeration that arise naturally in our model, even under the requirement of bilateral

proofness. The convergent paths considered are different, depending on the cost of linking

c. If c < 1/2, the process is illustrated in Figure 7. It consist of a series of adjustments by

which extremal players gradually close their distance to the center of the line, eventually

conforming a single star encompassing all players. In the alternative case where c > 1/2,

those aforementioned paths no longer embody bilateral-proof adjustments and must be

replaced by adjustments in which the lines stretching to either side of the central player

become progressively shorter by second-neighbors of that central player establishing a

direct connection. This process is illustrated in Figure 8.

The former considerations show the fragility of the cycle to the removal of a single link.

Now we argue that the star is a much more robust architecture, in that no single per-

turbation can trigger an adjustment process away from it. Commence with the star and

suppose that a link between the center, say player 1, and some other player i is deleted.

If player i then receives an opportunity to link to player 1, both players want to form the

link and the star is restored. So suppose that there is a linking opportunity between i

and another peripheral player j. Then, both will want to form the link as well and this

is a bilateral-proof adjustment. Of course, if player i next receives a linking opportunity

to 1, this link will be formed and the previously established link between i and j deleted

(since c > 1/6, which means that the original star is restored. Finally, consider the sit-

uation where i and some other peripheral player k receive a linking opportunity. It can

be checked that these players will not form a link if c > 5/12.9 Thus, in this case, no link

between i and any other peripheral player (different from j) will be created. Eventually,

i and the central player 1 will receive a linking opportunity, in which case the star will

restored.

9If c ≤ 5/12, the link between i and k would be formed and a hybrid cycle-line network would bereached that is nevertheless not stationary for the adjustment dynamics. In fact, based on considerationsused in Step 4 of our main Theorem, it can be shown that the cost is so low that the cycle may expandsteadily until it encompasses the whole population. In this case, therefore, one expects that, in thepresence of perturbations, the network architecture should switch indefinitely between the cycle and thestar.

20

The above discussion shows that a star is relatively more robust than a cycle. It is how-

ever silent on whether, and how, a non-empty network might be formed through gradual

adjustment when the linking cost c > 1/2. In general, this would require that the pertur-

bations/mutations also affect link creation, thus generating a sufficient “critical mass.”

A natural modelling option, for example, would be to posit that not only existing links

are destroyed with some small (and say, independent) probability ε > 0, but that every

non-existent link may arise under the same conditions. The adjustment process would

then become ergodic, thus guaranteeing the existence of a unique invariant probability

measure summarizing its long-run behavior. Following standard evolutionary literature,

the aim would be to identify the so-called stochastically stable states that arise with sig-

nificant long-run probability as ε ↓ 0. We conjecture that in such a dynamic model stars

would be uniquely stable for a range of cost values 1/6 < c < c.

5 Conclusion

This paper has studied a simple model of network formation where agents may exploit

positional advantages if these provide them with the ability to block profitable bilateral

interaction between two players who are not direct neighbors. We show that the strategic

struggle for these advantages leads to a polarized star architecture where a single player

becomes essential to connect every other pair of players. This represents a clear cut

formalization of the notion found in the sociological literature that structural holes opens

the potential for large benefits to those individuals who succeed in bridging them.

Appendix

The surplus bargaining game

Consider any pair of players, i and j, which may generate a unit of surplus (i.e. have at

least a network path joining them). Given the prevailing network g, the considerations

explained in the text induce a coalitional-form with transferrable utility given by the

characteristic function v : 2N → Rn defined, for each S ⊂ N as follows:

v(S) =

{1 if ∃{i1, . . . , in} ⊂ S s.t. gi,i1 · gi1,i2 · gi2,i3 · · · gin,j = 10 otherwise

. (3)

21

Given any imputation z ∈ Rn define the excess payoff that can be earned by any coalition

S by:

U(S, z) = v(S)−∑i∈S

zi

and the excess payoff that can be earned by some player k against other player l by:

uk`(z) = max{U(S, z) : S ⊂ N, k ∈ S, ` /∈ S}.

Then the Kernel of the game induced by the network g (which, in this case, also happens

to be in the Core10) is defined as the set of imputations z that satisfy, ∀k, ` ∈ N, one of

the following two conditions:

uk`(z) ≥ u`k(z); (4)

uk`(z) < u`k(z) ⇒ zk = v({k}) = 0. (5)

The intuitive basis for (4)-(5) is the idea that each player evaluates any imputation z

contemplated throughout the bargaining process on the basis of the induced excess payoffs

uk`(z). Thus, if there is a bilateral “imbalance” in these magnitudes for some pair of

players, this is sure to trigger a “reasonable objection” from the unfavored party (i.e. the

agent with the higher excess) unless the other is already at her minimum (individually

rational) payoff.

We now argue that, given the characteristic function (3) associated to a particular pair of

players i and j, the induced Kernel consists of the unique imputation vector z satisfying:

zk =

{ 1e(i,j)+2

if k ∈ E(i, j) ∪ {i, j}0 otherwise.

thus inducing the payoff function specified in (1). This conclusion follows from the fol-

lowing two claims.

Claim 2 Consider any pair of players k, ` ∈ E(i, j)∪{i, j}. Then, any Kernel imputation

z satisfies zk = z`.

Claim 3 Consider any player k /∈ E(i, j)∪{i, j}. Then, any Kernel imputation z satisfies

zk = 0.

10It is easy to check that the Core of the game associated to the surplus generated by i and j consistsof all those imputations where inessential players receive a null share.

22

To show Claim 2, suppose that k, ` ∈ E(i, j) ∪ {i, j} but zk > z`. Given that for any

S ⊂ N\{k} and for any S ′ ⊂ N\{`}, we have v(S) = v(S ′) = 0. Therefore,

uk`(z) = max{U(S, z) : S ⊂ N, k ∈ S, ` /∈ S} = U({k}, z) = −zk

< −z` = U({`}, z) ≤ max{U(S, z) : S ⊂ N, k` ∈ S, k /∈ S} = U`k(z).

By virtue of (5), this requires that zk = 0, which is a contradiction with the fact that

z` ≥ 0.

Next, to establish Claim 3, suppose that k /∈ E(i, j) ∪ {i, j} but zk > 0. Consider then

some other ` ∈ E(i, j) ∪ {i, j}. Again, since v(S) = 0 for any S ⊂ N\{`}, we have that

uk`(z) = −zk < 0.

Considering now the reciprocal excess payoff u`k(z), note that S0 ≡ E(i, j) ∪ {i, j} ⊂N\{k} so that v(S0) = 1 and, therefore,

u`k(z) = max{u(S, z) : S ⊂ N, k ∈ S, ` /∈ S}≥ u(S0, z) ≥ 1−

∑

u6=k

zu = zk > 0.

Therefore, (5) requires that zk = 0, a contradiction. ¥

Proof of Proposition 1: The proof of the result requires two preliminary lemmas. The

first one concerns critical links, i.e. links that define the only path between the two end

players (and whose deletion, therefore, would increase the number of components). It

establishes that the marginal payoff of any such critical link is equal for the two players

involved. The second lemma shows that in any component of a network, there are at

least two non-essential players, i.e. players who are not essential for any interaction (and

therefore enjoy only access payoffs).

Lemma 1 Consider any network g. If gi,j = 1 and the link is critical then Mi(gi,j; g) =

Mj(gi,j; g).

Proof: We show that the number By hypothesis gi,j is critical, and so it follows that i and

j lie in different components in the network g − gi,j. Let Ci(g), Cj(g) be the components

that contain i and j, respectively, where we shall usually dispense with an explicit account

of the dependence and simply write Ci and Cj. The marginal payoff of the link ij for

player i, is given by:

23

Mi(gi,j; g)

=1

2+

∑

k∈Cj\{j}

1

e(i, k) + 2+

∑

l∈Ci\{i}

∑

k∈Cj\{j}

1

e(l, k) + 2+

∑

l∈Ci\{i}

1

e(l, j) + 2− c

where the first two terms refer to access benefits while the latter two terms refer to

essentiality benefits. Similarly, we can write the marginal payoffs of player j form link gi,j

as:

Mj(gi,j; g)

=1

2+

∑

l∈Ci\{i}

1

e(j, l) + 2+

∑

l∈Ci\{i}

∑

k∈Cj\{j}

1

e(l, k) + 2+

∑

k∈Cj\{j}

1

e(i, k) + 2− c

It follows then that Mi(gi,j; g) = Mj(gi,j; g) and the proof of the Lemma is complete.

Lemma 2 In a network g, a component with m players has at least 2 non-essential

players.

Proof: Consider any arbitrary component of the network with m nodes. First, we note

that from any arbitrary connected network one can reach a line network by a series of

steps that involve only two operations: (i) removal of links; (ii) reconnection of existing

links to extremal players (i.e. players with only one link). It is easy to see that any

such operation can (weakly) increase the number of essential players. Since in the line

network the two extremal players are non-essential, it follows that the maximum number

of essential players in the original component can be no higher than m − 2. Thus any

component in a network with m players must have at least 2 non-essential players.

Equipped with Lemmas 1 and 2, we proceed with the proof of the proposition. Let

g be a non-empty BE network, and suppose it is not connected. Let C be the largest

component in g, which must therefore contain at least 2 players. We claim that there

is a player j /∈ C that can establish a mutually profitable link with some player in C.

For simplicity, we shall consider the extreme (and less favorable) case where j has no

connections, i.e. defines a singleton component.

By Lemma 2, we know that C has some non-essential player i ∈ C. Two possibilities

need to be considered separately. One is that i is extremal, i.e. she has only one link that

24

connects her to some other player ` in the component. Then, it is clear that, since i and

` both find it profitable to keep their link, player ` would find it optimal to create a link

with j if given the opportunity, and so would player j. This contradicts that the network

g may be a BE network.

The second possibility is that i is non-extremal and therefore ηi ≥ 2. Let Nmi (g) =

|{j ∈ Ci : e(i, j) = m)}| be the players whom i accesses via m essential players and let

ηmi (g) = |Nm

i (g)|. The payoffs of this player i in network g are then given by:

η0i (g)

2+

η1i (g)

3+

η2i (g)

4+ .... +

ηri (g)

r + 2− ηi(g)c (6)

for some r ≤ n− 2.

Since g is an equilibrium, it follows then

1

ηi(g)

[η0

i (g)

2+

η1i (g)

3+

η2i (g)

4+ .... +

ηri (g)

r + 2

]≥ c. (7)

Now let us examine marginal returns for player j /∈ C from a link with player i. Suppose,

for simplicity, that player j is a singleton component. Then the marginal returns to j are

given as follows:

η0j (g + gi,j)

2+

η1j (g + gi,j)

3+

η2i (g + gi,j)

4+ .... +

ηri (g + gi,j)

r + 2− c, (8)

where, by analogy with previous notation, g + gi,j simply denotes the network obtained

by replacing gi,j in network g by a new gi,j = 1.

Note now that for every k ∈ C\{i}, e(j, k; g +gi,j) = e(i, k; g)+1. Thus ηmj (g +gi,j) =

ηm−1i (g) for every m ≥ 1 and η0

j (g +gi,j) = 1. Using these facts we can write the marginal

returns of player j from the link with i as follows:

1

2+

η0i (g)

3+

η1i (g)

4+ .... +

ηri (g)

r + 3− c. (9)

Now we argue that

25

1

2+

η0i (g)

3+

η1i (g)

4+ .... +

ηri (g)

r + 3

>1

2

[η0

i (g)

2+

η1i (g)

3+

η2i (g)

4+ .... +

ηri (g)

r + 2

]

≥ 1

ηi(g)

[η0

i (g)

2+

η1i (g)

3+

η2i (g)

4+ .... +

ηri (g)

r + 2

]

≥ c

The first inequality is immediate, while we use ηi(g) ≥ 2 in deriving the second

inequality and equation (7), in deriving the final inequality. We now apply Lemma 1 to

conclude that player i also has a strict incentive to form a link with j, given that all

existing links are retained. But note that given that link gi,j is formed, player i has no

incentive to delete any of his erstwhile links since (roughly speaking) the marginal returns

from each of these links has actually increased. Thus players i and j have a strict incentive

to form an additional link. These arguments extend directly to cover the case where j

belongs to a non-singleton component. Thus g is not a BE network, a contradiction that

completes the proof. ¥

Proof of Theorem 1: As indicated in the text, the proof can be decomposed into five

steps. In what follows, each of these steps is formally embodied by a corresponding lemma.

All of them assume that c > 1/6 and n ≥ n(c) for some suitable n(c).

Lemma 3 The star is the only minimal network which can be sustained in a SBE.

Proof: Consider a g that is minimally connected but not a star. Then, there are two

players, say i and j, such that (a) E(i, j; g) ≥ 2; (b) they are “end players,” i.e. ηi(g) =

ηj(g) = 1. Then, it readily follows that for at least one of them, say i, the following holds:

There is a player x with e(i, x; g) = 1 (thus, x is two steps away from i in g) and the set

of players k for whom x is essential in connecting i and k has a cardinality that is at least

(n− 4)/2, i.e. |{k ∈ N : x ∈ E(i, k; g)| ≥ (n− 4)/2.

We argue that such a network g cannot be induced by an equilibrium. First note that,

given the linking cost c, the number of essential players that can be supported in equi-

librium between any two players has some (finite) upper bound e(c), independent of n.

Consider then the possibility that i and x were to form a link. The gross gains, ∆πi and

26

∆πx, induced by that change for i and x (if all other links were to remain in place) is

bounded below as follows:

min{∆πi, ∆πx} ≥ (n− 4)/2

e(c)− 1− (n− 4)/2

e(c)=

n− 4

2e(c)(e(c)− 1).

This expression is larger than c if n is large enough, which implies that both i and x

benefit from a deviation that creates a link between them and keeps all other links.

Lemma 4 There can be at most one cycle in a SBE network.

Proof: Suppose g is an equilibrium network and there are two or more cycles in it. Let

χ1 = (i1, i2, ...in) be ordered set of players in one cycle, and let χ2 = (j1, j2, ...jm) be those

in the other cycle. Since g is connected it follows that there are two possibilities: (1)

cycles have common players and (2) cycles have no common players. We take these up in

turn.

(1). Cycles have players in common: If there is a single common player i1 in the two cycles

then it is easy to see that the partners of i1 (say) i2 ∈ χ1 and j2 ∈ χ2 have a strict incentive

to delete their links with i1 and instead form a link with each other. (Throughout, we

shall abuse notation and write i ∈ χ if i is one of the nodes in the ordered collection of

nodes specifying χ.) Consider next the case with two or more players in common. Let

(i1, i2, ...ik) be the players in common. Suppose that k ≥ 3; the case of k = 2 is simple

and omitted. Then there exist players i1, ix, and jy with the following properties: ix ∈ χ1

but ix /∈ χ2, while jy ∈ χ2 and jy /∈ χ1 and gi1,ix = gi1,jy = gi1,i2 = ... = gik−1,ik = 1. Note

also that like player i1, ik must again have links with a player who belongs to one of χ1

and χ2 only. It then follows that ik−1 and jy have at least a weak incentive to delete their

current link with ik and i1 respectively and instead form a link with each other. It then

follows that g cannot be sustained by a SBE.

(2). Cycles have no common players. Since g is a SBE network it is connected and so

there exists a path between the two cycles. Let (i1, i2, ..., ik) be members of such a path

with i1 ∈ χ1 while ik ∈ χ2. Suppose gi1,ix = 1 and gik,jy = 1, where ix ∈ χ1 and jy ∈ χ2.

Now it is easy to use a variant of the earlier argument for case 1 above to show that

players ix and jy have a strict incentive to delete their link with i1 and ik and instead

form a link with each other. The proof is complete. ¥

27

Lemma 5 A cycle containing all players cannot be sustained in a BE.11

Proof: Consider two player i and j who are furthest apart in terms of geodesic distance

in the cycle. Now consider the deviation in which each of the players deletes one link

and they form a link with each other in such a way that they create a line. Assume, for

simplicity, that n is even, so that there are (n − 2)/2 players to one side of player i and

(n − 2)/2 players to the other side of player j in the line created. We now show that

players i and j will strictly increase their payoff with this coordinated deviation.

We proceed in two steps: the first step is to show that individual payoffs are strictly

increasing as we move toward the center of the line. The payoffs of an individual player

consist of two components, the returns from accessing others and the returns from being

essential on paths between pairs of other players. Number the players on a line as 1, 2,

...n. The access returns to player l are given by

1

l+

1

l − 1+ ... +

1

2+

1

2+ ... +

1

n− l + 1(10)

while the access returns to player l + 1 are given by

1

l + 1+

1

l+ ...

1

2+

1

2+ ... +

1

n− l(11)

It now follows that access returns for player l +1 are larger than access returns for player

l if l < n/2.

We now turn to the returns from being essential. The essentialness payoff to player l can

be written as follows:

l−1∑i=1

n∑

j=l+2

1

e(i, j) + 2+

l−1∑i=1

1

e(i, l + 1) + 2(12)

Similarly, the essentialness payoffs to player l + 1 can be written as follows:

l−1∑i=1

n∑

j=l+2

1

e(i, j) + 2+

n∑

j=l+2

1

e(l, j) + 2(13)

The first part of the essentialness payoffs to the two players are equal, while the second

part of the payoffs are greater for player l + 1 if l < n/2.

11Here, we need that n(c) ≥ 4, since for n = 3 a complete network can be sustained in equilibrium forc < 1/6.

28

Let gC and gL denote the networks prevailing before the contemplated deviation and after

it (i.e. the cycle and the line with i and j at the center, respectively). To show that i and

j indeed obtain higher payoffs under gL, note that the aggregate gross payoffs obtained in

both cases are the same. The above argument implies that i and j enjoy a higher share

of total gross value in the line as compared to the other players. This implies that players

i and j earn a higher gross payoff in the line. Since their linking cost is the same in both

cases (i.e. 2c), it follows that they obtain higher net payoffs as well, which completes the

proof. ¥

Lemma 6 A SBE network with a cycle has the hybrid star-cycle architecture, for large

enough n.

Proof: Suppose that g is a non-empty SBE network with a cycle. Let X be the set of

players and x the number of players who are outside the cycle, while Y is the set of players

in cycle and define y = n − x to be number of players in cycle. Note that y ≥ 4 since

y = 3 cannot be sustained in equilibrium given that c > 1/6. Next we argue that, for

any such c, there is a y(c) such that y ≤ y(c) in equilibrium. Given step 2 above, clearly

x ≥ 1. Suppose gi,j = 1 for some i ∈ X and j ∈ Y . Since j ∈ Y , there is some k ∈ Y such

that gj,k = 1. Clearly, player k prefers strictly to switch link from j to i. This reduces the

essentialness payoffs he has to pay out to j and keeps his costs constant. On the other

hand, i benefits from such an adjustment if (y − 1)/6 > c. So given a c, in order for the

cycle to be stable, we must have y ≤ y(c) ≡ 6c + 1.

Next we show that all nodes that do not belong to the cycle must be connected to a single

node in the cycle. The initial observation is that, if n is large enough, there cannot be

two distinct trees with different roots lying in the cycle. To see this, consider one of those

subtrees that has a number of nodes at least equal to [n− y(c)] /y(c) = (n/y(c))− 1. At

least one such tree must exist since there are at most y(c) nodes in the cycle. Let i1 be

the root of this tree and i2 be the root of any other cycle. Now consider any node j in

the second tree different from its root, i2. A straightforward adaptation of the arguments

used in the proof of Lemma 3 lead to the conclusion that, if n is large, j and i1 can both

profit from forming a link, whether or not j maintains its link with i2. The reason is that,

since the number of essential players e(`, `′) for any pair of players, ` and `′, is bounded

by some e(c), independently of n, the gross gains from the link between j and i1 grow

29

linearly with n for both players, independently of whether player j maintains the link to

i2.

Once established that there can be at most one tree connected to the cycle through its

root, again relying on the arguments introduced in Lemma 3 we arrive at the conclusion

that, for any two nodes in the tree, u and v, the number of essential players e(u, v) ≤ 1.

This still leaves open the possibility that the tree consists of a star with a center at some

node ic that is not part of the cycle but has a link to a node in the cycle. But, in that case,

the node ic and any of the neighbors of k in the cycle, say k′, both profit from establishing

a direct link, for large enough n. ¥

Lemma 7 The star is the unique hybrid-cycle equilibrium.

Proof: As argued in the previous step, y ≤ y(c) for a function y(c) that is independent

of n. Thus, fixing some c, consider the class of hybrid networks g in which y ≤ y(c). Let

i be the player in the center of the star and suppose that j, k ∈ Y with gi,j = gi,k = 1.

We now show that j and k have a strict incentive to form a link if number of peripheral

players x ≥ n − y(c) is sufficiently large. The payoffs of players j and k in the hybrid

network g are given by

πj(gH) = πk(g

H) =x

3+

y − 1

2− 2c (14)

Now consider a deviation by players j and k in which player k deletes his link with

player i and player j deletes his link with player m in the cycle and instead players j and

k form a link with each other. The resulting network is a minimal network g′ in which

there are x peripheral players and a line starting with player i which consists of y players.

The payoffs of player j in g′ are given by

πj(g′) =

x

3+

1

2+

y−1∑

k=2

1

k+ x

y+1∑

k=4

1

k+

y∑

k=3

1

k− 2c (15)

These payoffs are bounded below by

x

3+ x

y − 2

y + 1− 2c = M (16)

Next note that M > x/3 + (y − 1)/2− 2c if

x >y − 1

2

y + 1

y − 2(17)

30

Since y ≥ 4, the right hand side is increasing in y and bounded above by [y(c) −1][y(c) + 1]/2[y(c) − 2]. The final step is to note that x = n − y ≥ n − y(c) and so (17)

applies for sufficiently large n. Thus player j has a strict incentive to switch links to

player k for large n. We now turn to the incentives of player k.

The payoffs of player k in g′ are given by

πk(g′) =

x

4+

1

3+

1

2+

y−2∑

k=2

1

k+ x

y+1∑

k=5

1

k+

y∑

k=4

1

k− 2c (18)

These payoffs are bounded below by

x

4+ x

y − 3

y + 1− 2c = M ′ (19)

Note that M ′ > x/3 + (y − 1)/2− 2c if

x :>:6(y − 1)(y + 1)

11y − 37(20)

Since y ≥ 4, the right hand side is positive and increasing in y and so is bounded

above by 6[y(c)− 1][y(c) + 1]/[11y(c)− 37]. Note that x = n− y ≥ n− y(c) is larger than

this term for sufficiently large n. Thus player k has a strict incentive to switch links to

player j, for sufficiently large n.

Combining Lemmas 3-7 the proof of Theorem 1 is complete. ¥

Proof of Proposition 3: In a cycle, all players are symmetrically located, so we need

only consider the incentives for a typical player, say some player i. First note that the

payoffs of player i in a cycle are (n− 1)/2− 2c. The payoffs from deleting both links are

0 and clearly for a given c, there is always a large enough n such that deleting both links

is not optimal. Consider next the deviation of deleting one link. If player i deletes one

link and keeps the other link as in the cycle then he becomes an end-player of the line

network. In this network his payoffs is given by

n∑

k=2

1

k− c (21)

The payoff from both links in cycle are higher if

n− 1

2−

n∑

k=2

1

k> c (22)

31

Clearly this inequality holds for large enough n. We next consider the deviation in which

player i deletes both links but forms a new link (say) with the center of the line network

that arises. Again, a variation of the above argument shows that for large n this reduces

payoffs.

We finally consider deviations by player i which involve coordinated deviation with one

other player. Suppose that in the cycle he has a link with i−1 and i+1. In the deviation,

he maintains the link with i − 1 but deletes the link with i + 1 and instead forms a link

with some player k. If player k retains all his links as in the cycle, it is easy to see that the

payoffs of the player go down strictly since the costs remain the same (he maintains two

links) while the gross payoffs decline since player k is essential for accessing at least one

player, namely i + 1. So this deviation is not profitable. If player k deletes both his links

then the payoffs from the deviation are still lower and so it clearly not profitable. We

turn to the final case, where players i and k coordinate and link up with each other but

delete a link each so that the new network is a line. From Lemma 5 it follows that players

such a deviation is profitable. However, now we need to check whether this deviation is

credible: do the players have an incentive to actually delete one of their links? We show

that for large n, at least on the players has a strict incentive to retain their links in the

cycle. Number the players in the line as 1, 2...., n, from left to right. So there are i − 1

players to the left of player i while there are n − k players to the right of player k in

the line. Suppose without loss of generality that i − 1 ≥ n − k. Player k gets a payoff∑i+1l=2 1/l in the line network from these players 1, 2..., i. The payoffs can be increased to

i/2 if player k forms a link with player 1. Clearly, it is profitable for player k to deviate

from the deviation and form a link with player 1, if n is large. It is similarly possible to

show that player 1 would have an incentive to form a link with player k, if n is large.

Thus the deviation in which i and k deviate to create a line is not two-person coalition

proof. Finally, we note that starting from a cycle it is not profitable for player i to form

any additional links. The proof for cycle being a SBPE network is complete. ¥

Remark 1 Hybrid cycle-star networks are SBPE

Consider a hybrid cycle-star network with x nodes out of the cycle and y nodes in the

cycle. We know from the Lemma 6 that, in order for such a configuration to be an

equilibrium, it must be that y ≤ 6c + 1. Under the latter inequality, it is clear that no

(bilateral) deviation involving a player outside the cycle can be profitable for this player.

32

Thus, we are left with deviations involving pairs of non-neighboring players in the cycle,

as those considered in Lemma 7. Let i and j be any such players. The only way in which

they can gain is by attempting to become “central.” This must involve establishing a link

between them and, simultaneously, destroying one of their respective links to the cycle.

Now we argue that any such deviation is not immune to a unilateral omission by at least

one of the players in her link deletion. By keeping her link, player i can access one half

of the players in the cycle without any intermediation costs. Thus, for large y the gain

∆Πi from retaining the link (assuming j abides by the deviation) can be approximated

as follows:

∆Πi ' 1

2

y

2− ln

y

2− c.

Now suppose y = c/α for some α ∈ (1/6, 1/4). Then y < 6c + 1 but ∆Πi > 0 for large

enough c and y. Similarly, it is possible to show that a member of the cycle has no

incentives to delete links with cycle members and switch to a link with the center of

star. These computations put together show that there is a SBPE that supports a hybrid

cycle-star network. ¥

Remark 2 The cycle is not dynamically robust

Starting from a line network we claim that there is positive probability that the ad-

justment dynamics converges to a star network. We need to consider separately two cases:

c < 1/2 and c ≥ 1/2.

Case 1: c < 1/2: After the deletion of the link, the resulting network is a line, so

let players be indexed consecutively from one end to the other, i = 1, 2, ...., n, with n

odd for simplicity. Consider the following inductive argument. Assume that, at some

stage of the process, the network has, on the one hand, the players 1, 2, ..., m − 1 (m ≤(n+1)/2) displaying a single link that connects them to player m and, on the other hand,

the players r + 1, ..., n (r ≥ (n + 1)/2) with a single link connecting them to player r.

Furthermore, suppose that players j = m,m + 1, ..., r are all connected through a single

path. We claim that, with positive probability, a suitable chain of bilateral adjustments

may bring the network to a situation such as the starting one with the indices of players

m and r replaced, respectively, by m′ = m + 1 and r′ = r − 1. This inductive procedure

stops when m = r, in which case a star network already prevails.

33

To see this, suppose first that all players i = 1, 2, ...m − 1 receive in turn a bilateral

revision opportunity with player m + 1. We argue that, if the population is large, a new

link between i and m+1 will be formed, irrespectively of whether or not these players want

to delete as well some of their other links. The reason is that, keeping their current links,

the fresh link to be created generates an additional benefit to the two players involved (i

and m + 1) that can be bounded below by a number arbitrarily close to 1/2 when n is

large. Specifically, the gross marginal benefit ∆Π for either player of connecting directly

(rather than through player m) is bounded as follows:

∆Π ≥ (1

2− 1

3) + (

1

3− 1

4) + (

1

4− 1

5) + ... + (

1

n−m + 1− 1

n−m + 2)

=1

2− 1

n−m + 2

≥ 1

2− 2

n + 3

which is indeed larger than c (< 1/2) for large enough n. This implies that by creating

(and paying for) a link, players i and m + 1 can benefit, even if they hold fixed all other

links they currently support. If any one or them were to simultaneously destroy some of

their other links, the marginal profitability of the link in question can only increase. Thus,

if i deletes her (only) link to m, then m + 1 becomes the unique (and therefore “most

valuable”) partner of i, whereas if m+1 denotes any subset of her other links, the marginal

gains of the link can only increase as well. In sum, the adjustment is bilateral-proof.

After this round of bilateral revisions is completed, there is a link between every i =

1, 2, ...,m − 1 and m + 1. But this implies that none of the links between i and m are

profitable for the latter player if c > 1/6. Thus, they will be removed if a further round of

revision is given to the player m (say, as part of a bilateral adjustment between this player

and m + 1). At this point, we arrive at a network where all players i = 1, 2, ..., m have

their links as desired. But, clearly, the same considerations may be applied to players

i = r, r + 1, ..., n, which establishes the inductive argument.

Case 2: c ≥ 1/2: Again, after the deletion of the link, the resulting network is a line,

so we continue to index players consecutively from one end to the other, i = 1, 2, ...., n,

with n odd for simplicity. Suppose that player n−32

is given the option of linking to the

central player in the line, n+12

and deleting her link to player n−12

. Both players strictly

benefit from this adjustment if the population is large enough. Concerning player n−32

, this

happens because the payoffs she obtains from players n−12

and n+12

are simply permuted,

34

whereas in the modified network the number of essential players that n−32

requires to

access each i = n+32

, n+52

, ..., n is one less. On the other hand, to see that player n+12

also

benefits strictly from the adjustment note that her gain in gross payoffs ∆Π is given by:

∆Π =

[(1

2− 1

3) + (

1

3− 1

4) + ... + (

1n−1

2− 1

− 1n−1

2

)

]

+

[(

1

2 + 1− 1

3 + 1) + (

1

3 + 1− 1

4 + 1) + ... + (

1n−1

2− 1 + 1

− 1n−1

2+ 1

)

]+ ...

+

[(

1

2 + n−12

− 1

3 + n−12

) + (1

3 + n−12

− 1

4) + ... + (

1n−1

2− 1 + n−1

2

− 1n−1

2+ n−1

2

)

]

=

[1

2− 1

n−12

]+

[1

2 + 1− 1

n−12

+ 1

]+

[1

2 + n−12

− 1n−1

2+ n−1

2

]

=

n2∑

r=0

[1

2 + r− 1

n−12

+ r

].

As n grows, the asymptotic behavior of the above expression can be approximated by

∫ n2

0

[1

2 + r− 1

n−12

+ r

]dr = ln(2 +

n− 1

2)

which increases unboundedly as n ↑ ∞. The net gains of the contemplated adjustment

for player n+12

consist of the gross gain ∆Π net of the cost c for the additional link she

supports with n−32

. It follows, therefore, that, given any c, the net gains ∆Π − c > 0 as

long as n is large enough.

However, in connection to the previous bilateral adjustment by players n−32

and n+12

, the

question remains as to whether it is bilateral proof. Clearly, no deviation from such an

adjustment can be optimal for player n+12

. On the other hand, for player n−32

, it is still

conceivable that she might want to deviate and maintain the link to n−12

, since this saves

the need to pay intermediation costs to player n+12

when player n−12

accesses n−32

and all

players i < n−32

. The gross gain derived from such a deviation would be:

∆Π = (1

2− 1

3) + (

1

3− 1

4) + (

1

4− 1

5) + ... + (

1n−1

2

− 11

n+1

)

=1

2− 2

n + 1,

35

which is always lower than c (≥ 1/2). Thus, such a deviation is not profitable and, conse-

quently, the contemplated adjustment is bilateral-proof. Now we may proceed iteratively

with players that are two steps apart from the central player in the current network (e.g.

either player n−52

or n+32

at the next iteration). It is immediate to see that none of the for-

mer considerations concerning the profitability and bilateral-proofness of the adjustment

until a star centered at player n+12

is finally reached. At this point, no further adjustment

possibilities will change the network. ¥

References

[1] Albert, R. and A.-L. Barabasi (2002): Statistical mechanics of complex networks,

Review of Modern Physics , 74, 47-97.

[2] Ahuja, G. (2000), Collaboration networks, structural holes, and innovation: A logi-

tudinal study, Administrative Science Quarterly, 45, 3, 425-455.

[3] Aumann, R. and R. Myerson (1988): Endogenus formation of links between players

and coalitions: an application to the Shapley Value, in The Shapley Value, A. Roth

(ed.), 175-91, Cambridge University Press.

[4] Bala, V. and S. Goyal (1999): A Non-Cooperative Model of Network Formation,

Econometrica , 68, 1181-1230.

[5] Barabasi, A.-L. and R. Albert (1999): Emergence of scaling in random networks,

Science 286, 509-12.

[6] Boorman, S., (1975), A combinatorial optimization model of transmission of job

information through contact networks. Bell Journal of Economics, 6, 216-49.

[7] Burt, R. S (2004), Structural holes and good ideas, American Journal of Sociology,

110, 349-399

[8] Burt, R.S. (1994): Structural Holes, New York: Academic Press.

[9] Calvo, A. and M. Jackson (2004), The effects of social networks on employment and

inequality, American Economic Review, forthcoming.

[10] Coleman, J., E. Katz and H. Menzel (1966), Medical Innovation. Bobbs-Merrill. New

York.

36

[11] Davis, M. and M. Maschler (1965): The Kernel of a cooperative game, Naval Research

Logistics Quarterly, 12, 223-259.

[12] Dutta, B., A. van Den Nouweland and S. Tijs (1995): Link formation in cooperative

situations, International Journal of Game Theory 27, 245-256.

[13] Ellison, G. and D. Fudenberg (1992), Rules of thumb for social learning, Journal of

Political Economy, 101, 612-644.

[14] Feri, F. (2004), Stochastic stability in networks with decay, mimeo, University of

Venice.

[15] Goyal, S. and J.L Moraga (2001), R&D Networks Rand Journal of Economics, 32,

4, 686-707.

[16] Granovetter, M. (1974): The Strength of Weak Ties, American Journal of Sociology,

78 , 1360-1380.

[17] Greif, A. (1994): On the Political Foundations of the Late Medieval Commercial

Revolution: Genoa during the Twelfth and Thirteenth Centuries, The Journal of

Economic History, 54, 271-287.

[18] Gulati, R., N. Nohria, and A. Zaheer (2000): Guest editor’s introduction to the

special issue: strategic networks, Strategic Management Journal 21, 199-201.

[19] Galeotti, A., and M. Melendez (2004), Exploitation and cooperation in networks,

Tinbergen Institute Discussion Paper 04-076/1.

[20] Hojman, D., and A. Szeidl (2004), Core and periphery in endogenous networks,

mimeo, Department of Economics, Harvard University.

[21] Jackson, M. O. (2003), The stability and efficiency of economic and social networks,

on Networks and Groups: models of strategic formation, eds. B. Dutta and M. O.

Jackson. Springer Verlag. Berlin.

[22] Jackson, M. O. and A. Wolinsky (1996): A Strategic model of economic and social

networks, Journal of Economic Theory, 71, 44-74.

[23] Kranton, R. & D. Minehart (2001): A theory of buyer-seller networks, American

Economic Review 91, 485-508.

37

[24] Mehra, A. M. Kilduff and D. Bass (2003), The social networks of high and low self-

monitors: implications for workplace performance, Administrative Science Quarterly,

46, 121-146.

[25] Montgomery, J. (1991), Social networks and labor-market outcomes: toward an eco-

nomic analysis, American Economic Review, 81, 5, 1408-1418.

[26] Newman, M. E. J. (2001): The structure of scientific collaboration networks, Proc.

Natl. Acad. Sci. USA, 98, 409-415.

[27] Podolny, J and J. Baron (1997), Resources and relationships, social networks and

mobility in the work place, American Sociological Review, 62, 673-693.

[28] Rauch, J. E. (2001): Business and Social Networks in International Trade Journal of

Economic Literature 39, 1177-1203.

[29] Rees, A. (1966), Information networks in labour markets, American Economic Re-

view, 56, 559-566.

[30] Valente, T. (1996): Social Network Thresholds in the Diffusion of Inovations, Social

Networks 18, 69-89.

[31] Vega-Redondo, F. (2005): Diffusion, Play, and Search in Complex Social Networks,

forthcoming in the Econometric Society Monograph Series, Cambridge University

Press.

38

A. Star B. Cycle

C. Hybrid cycle-star

Figure 1: Main Architectures

2 3

A. Minimal network with 2 essential players

1

2

3 B. Players 1 & 2 deviate yielding a star

1

Figure 2: Incentives in minimal networks

3 4

I. Cycles with no common players

1 2

3 4

II. Players 3 and 4 deviate yielding one cycle1 2

Figure 3A: Instability of networks with two cycles

4

1

3

2

I. Network with two cycles

4

1

3

2

II. Players 1 and 2 deviate yielding one cycle

Figure 3B. Instability of networks with two cycles

11

22

A. Cycle B. A line

Figure 4: Bilateral deviations from cycle

2

1

A. Hybrid cycle-star

2

1B. Players 1 and 2 have an incentive to deviate

Figure 5: Instability of hybrid cycle-star

2

1

2

1

2

Bilateral deviation

Player 1 retains linkWith 3

C. Hybrid cycle-star

by 1 & 2

3

1 B. Line

A. Cycle

Player 1deviatesagain

Figure 6: Bilateral proof deviations

m

m

m+1

Figure 7: Process of agglomeration, c < 1/2

r

r+1

m+1

r

r+1

B. Network after moves

A. Initialnetwork Sequence

of bilateral-proofmoves

Sequence

of bilateral-proofmoves

m

Figure 8: Process of agglomeration, c > 1/2

r+1

A. Initialnetwork

B. Network after moves

r+1

m+1m

m+1

r

r